ReleaseBytes
anthropic Claude Platform Release Notes ·

Claude Platform Updates: Opus 5, Agent Enhancements, API Key Expiration

aipreview
feature patch breaking

Anthropic has released Claude Opus 5 with significant context window and output token improvements, alongside updates to Claude Managed Agents and API key management. These changes impact developers and architects using the Claude API, particularly those focused on advanced AI capabilities and agent development. Key features include a 1M token context window for Opus 5 and new webhook event types for agents, with some changes potentially requiring migration.

  • →Claude Opus 5 Launched with Extended Context and Output
  • →Opus 5 Breaking Change: Disabling 'Thinking' Effort Levels
  • →Mid-Conversation Tool Changes Now in Beta
  • →Fast Mode Removed for Claude Opus 4.7
  • →Effort Level Configuration for Claude Managed Agents
Breaking changes (1) ›
  • Opus 5 Breaking Change: Disabling 'Thinking' Effort Levels

    On Claude Opus 5, disabling the 'thinking' parameter is now restricted to 'effort: high' or below. Higher effort levels will result in a 400 error, a change from Claude Opus 4.8.

Deprecations (1) ›
  • Fast Mode Removed for Claude Opus 4.7

    Fast mode is no longer supported for Claude Opus 4.7. Requests made with 'speed: "fast"' will now return an error, with users advised to migrate to Opus 5 or Opus 4.8 for continued fast mode access.

Features (3) ›
  • Claude Opus 5 Launched with Extended Context and Output

    Claude Opus 5 is now available, featuring a 1M token context window and 128k max output tokens. It supports 'thinking on' by default and is priced identically to Opus 4.8.

  • Mid-Conversation Tool Changes Now in Beta

    The ability to add or remove tools during a conversation is now in beta for multiple Claude models, including Opus 5. This feature allows for dynamic tool management while preserving prompt cache.

  • Effort Level Configuration for Claude Managed Agents

    Users can now set an 'effort' level directly on a Claude Managed Agent's model configuration when creating it. This allows for finer control over agent behavior and capabilities.

Enhancements (1) ›
  • New 'Default' Mode for Fallback Parameter

    The 'fallbacks' parameter now supports a 'default' mode, which uses Anthropic's recommended models based on refusal category. This server-side fallback feature is currently in beta.

Read the original announcement →

https://platform.claude.com/docs/en/release-notes/overview

Related releases